We present open-source deep learning based classifiers for predicting physical therapy exercises using inertial measurement units, and a comprehensive analysis of impacts of sensor density, location, type, state estimation, and sample size on the performance.
GitHub: https://github.com/CMU-MBL/IMU_Exercise_Prediction.
Data: Downloads/View
Preprint: https://doi.org/10.36227/techrxiv.23296487.v1.
Data: Nineteen (19) subjects were recruited to perform 37 lower-body exercises while wearing ten (10) inertial measurement units (IMUs) on chest, pelvis, wrists, thighs, shanks, and feet. Check our preprint for more details of the data collection. Available data include:
1. IMU data (100 Hz).
2. Ideal joint angles of hips, knees, and ankles obtained from a marker-based motion capture system (100 Hz).
Code: Instruction can be found on the GitHub link above.
Citation: If you find this helpful for your project, please consider citing the following paper: Phan, Vu; Song, Ke; Silva, Rodrigo Scattone; Silbernagel, Karin G.; Baxter, Josh R.; Halilaj, Eni (2023). Seven Things to Know about Exercise Monitoring with Inertial Sensing Wearables. TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.23296487.v1.